Facundo Sapienza

This is a great discussion to have in person! And I agree we should agree on this since it is important for the structure of the packages. I still don't...

Hi @lesteve, thank you for your response. Weird you don't observe the problem when scaling X, but at least you see it when scaling Y. I agree that there is...

Hi @adam2392. I observe the issue with the three examples. I think the important point is that of course we expect numerical errors once this numbers became very small, but...

Hi @albangossard! Thanks for all the comments! Here some follow up: > In the specific setting where all components are of the same "nature" like when we are differentiating the...

Ok, I run some simple benchmark with different finite difference codes: ```julia res1 = @benchmark test_grad_finite_diff(ContinuousAdjoint(VJP_method=DiscreteVJP()); thres=[1.0, 1.0, 1.0], finite_difference_method = :Manual) res2 = @benchmark test_grad_finite_diff(ContinuousAdjoint(VJP_method=DiscreteVJP()); thres=[1.0, 1.0, 1.0], finite_difference_method...

Closed in #298

Sounds good! For reference, this is currenly implemented in the CI in [SphereUDE](https://github.com/ODINN-SciML/SphereUDE.jl), so we can use that repo for reference of how to to the automatic formatter.

I think I run into this issue... I left a post in the [Discourse](https://discourse.julialang.org/t/stop-condapkg-for-finding-dependencies-in-other-projects/122562) explaining my problem.